{-# language CPP #-} {-# language QuasiQuotes #-} {-# language TemplateHaskell #-} #if __GLASGOW_HASKELL__ >= 800 {-# options_ghc -Wno-redundant-constraints #-} #endif module OpenCV.Photo ( InpaintingMethod(..) , decolor , inpaint , denoise_TVL1 , fastNlMeansDenoisingColored , fastNlMeansDenoisingColoredMulti ) where import "base" Data.Int ( Int32 ) import "base" Data.Word ( Word8 ) import qualified "inline-c" Language.C.Inline as C import qualified "inline-c-cpp" Language.C.Inline.Cpp as C import "this" OpenCV.Internal.C.Inline ( openCvCtx ) import "this" OpenCV.Internal.C.Types ( withPtr ) import "this" OpenCV.Internal.Exception import "this" OpenCV.Internal.Photo.Constants import "this" OpenCV.Internal.Core.Types.Mat import "this" OpenCV.Internal.Core.Types ( withArrayPtr ) import "this" OpenCV.TypeLevel import qualified "vector" Data.Vector as V -------------------------------------------------------------------------------- C.context openCvCtx C.include "opencv2/core.hpp" C.include "opencv2/photo.hpp" C.using "namespace cv" -------------------------------------------------------------------------------- data InpaintingMethod = InpaintNavierStokes -- ^ Navier-Stokes based method. | InpaintTelea -- ^ Method by Alexandru Telea. deriving Show marshalInpaintingMethod :: InpaintingMethod -> Int32 marshalInpaintingMethod = \case InpaintNavierStokes -> c'INPAINT_NS InpaintTelea -> c'INPAINT_TELEA {- | Restores the selected region in an image using the region neighborhood. Example: @ inpaintImg :: forall h h2 w w2 c d . ( Mat (ShapeT [h, w]) ('S c) ('S d) ~ Bikes_512x341 , h2 ~ ((*) h 2) , w2 ~ ((*) w 2) ) => Mat ('S ['S h2, 'S w2]) ('S c) ('S d) inpaintImg = exceptError $ do maskInv <- bitwiseNot mask maskBgr <- cvtColor gray bgr maskInv damaged <- bitwiseAnd bikes_512x341 maskBgr repairedNS <- inpaint 3 InpaintNavierStokes damaged mask repairedT <- inpaint 3 InpaintTelea damaged mask withMatM (Proxy :: Proxy [h2, w2]) (Proxy :: Proxy c) (Proxy :: Proxy d) black $ \\imgM -> do matCopyToM imgM (V2 0 0) damaged Nothing matCopyToM imgM (V2 w 0) maskBgr Nothing matCopyToM imgM (V2 0 h) repairedNS Nothing matCopyToM imgM (V2 w h) repairedT Nothing where mask = damageMask w = fromInteger $ natVal (Proxy :: Proxy w) h = fromInteger $ natVal (Proxy :: Proxy h) @ <> -} inpaint :: (channels `In` [1, 3]) => Double -- ^ inpaintRadius - Radius of a circular neighborhood of each -- point inpainted that is considered by the algorithm. -> InpaintingMethod -> Mat ('S [h, w]) ('S channels) ('S Word8) -- ^ Input image. -> Mat ('S [h, w]) ('S 1) ('S Word8) -- ^ Inpainting mask. -> CvExcept (Mat ('S [h, w]) ('S channels) ('S Word8)) -- ^ Output image. inpaint inpaintRadius method src inpaintMask = unsafeWrapException $ do dst <- newEmptyMat handleCvException (pure $ unsafeCoerceMat dst) $ withPtr src $ \srcPtr -> withPtr inpaintMask $ \inpaintMaskPtr -> withPtr dst $ \dstPtr -> [cvExcept| cv::inpaint( *$(Mat * srcPtr) , *$(Mat * inpaintMaskPtr) , *$(Mat * dstPtr) , $(double c'inpaintRadius) , $(int32_t c'method) ); |] where c'method = marshalInpaintingMethod method c'inpaintRadius = realToFrac inpaintRadius {- | Perform fastNlMeansDenoising function for colored images. Denoising is not per channel but in a different colour space Example: @ fastNlMeansDenoisingColoredImg :: forall h w w2 c d . ( Mat (ShapeT [h, w]) ('S c) ('S d) ~ Lenna_512x512 , w2 ~ ((*) w 2) ) => Mat ('S ['S h, 'S w2]) ('S c) ('S d) fastNlMeansDenoisingColoredImg = exceptError $ do denoised <- fastNlMeansDenoisingColored 3 10 7 21 lenna_512x512 withMatM (Proxy :: Proxy [h, w2]) (Proxy :: Proxy c) (Proxy :: Proxy d) black $ \\imgM -> do matCopyToM imgM (V2 0 0) lenna_512x512 Nothing matCopyToM imgM (V2 w 0) denoised Nothing where w = fromInteger $ natVal (Proxy :: Proxy w) @ <> -} fastNlMeansDenoisingColored :: Double -- ^ Parameter regulating filter strength for luminance component. -- Bigger h value perfectly removes noise but also removes image -- details, smaller h value preserves details but also preserves -- some noise -> Double -- ^ The same as h but for color components. For most images value -- equals 10 will be enough to remove colored noise and do not -- distort colors -> Int32 -- ^ templateWindowSize Size in pixels of the template patch that -- is used to compute weights. -- Should be odd. Recommended value 7 pixels -> Int32 -- ^ searchWindowSize. Size in pixels of the window that is used -- to compute weighted average for given pixel. Should be odd. -- Affect performance linearly: greater searchWindowsSize -- - greater denoising time. Recommended value 21 pixels -> Mat ('S [h, w]) ('S 3) ('S Word8) -- ^ Input image 8-bit 3-channel image. -> CvExcept (Mat ('S [h, w]) ('S 3) ('S Word8)) -- ^ Output image same size and type as input. fastNlMeansDenoisingColored h hColor templateWindowSize searchWindowSize src = unsafeWrapException $ do dst <- newEmptyMat handleCvException (pure $ unsafeCoerceMat dst) $ withPtr src $ \srcPtr -> withPtr dst $ \dstPtr -> [cvExcept| cv::fastNlMeansDenoisingColored( *$(Mat * srcPtr) , *$(Mat * dstPtr) , $(double c'h) , $(double c'hColor) , $(int32_t templateWindowSize) , $(int32_t searchWindowSize) ); |] where c'h = realToFrac h c'hColor = realToFrac hColor {- | Perform fastNlMeansDenoisingColoredMulti function for colored images. Denoising is not pre channel but in a different colour space. This wrapper differs from the original OpenCV version by using all input images and denoising the middle one. The original version would allow to have some arbitrary length vector and slide window over it. As we have to copy the haskell vector before we can use it as `std::vector` on the cpp side it is easier to trim the vector before sending and use all frames. Example: @ fastNlMeansDenoisingColoredMultiImg :: forall h w w2 c d . ( Mat (ShapeT [h, w]) ('S c) ('S d) ~ Lenna_512x512 , w2 ~ ((*) w 2) ) => Mat ('S ['S h, 'S w2]) ('S c) ('S d) fastNlMeansDenoisingColoredMultiImg = exceptError $ do denoised <- fastNlMeansDenoisingColoredMulti 3 10 7 21 (V.singleton lenna_512x512) withMatM (Proxy :: Proxy [h, w2]) (Proxy :: Proxy c) (Proxy :: Proxy d) black $ \\imgM -> do matCopyToM imgM (V2 0 0) lenna_512x512 Nothing matCopyToM imgM (V2 w 0) denoised Nothing where w = fromInteger $ natVal (Proxy :: Proxy w) @ <> -} fastNlMeansDenoisingColoredMulti :: Double -- ^ Parameter regulating filter strength for luminance component. -- Bigger h value perfectly removes noise but also removes image -- details, smaller h value preserves details but also preserves -- some noise -> Double -- ^ The same as h but for color components. For most images value -- equals 10 will be enough to remove colored noise and do not -- distort colors -> Int32 -- ^ templateWindowSize Size in pixels of the template patch that -- is used to compute weights. Should be odd. -- Recommended value 7 pixels -> Int32 -- ^ searchWindowSize. Size in pixels of the window that is used to -- compute weighted average for given pixel. Should be odd. -- Affect performance linearly: greater searchWindowsSize - -- greater denoising time. Recommended value 21 pixels -> V.Vector (Mat ('S [h, w]) ('S 3) ('S Word8)) -- ^ Vector of odd number of input 8-bit 3-channel images. -> CvExcept (Mat ('S [h, w]) ('S 3) ('S Word8)) -- ^ Output image same size and type as input. fastNlMeansDenoisingColoredMulti h hColor templateWindowSize searchWindowSize srcVec = unsafeWrapException $ do dst <- newEmptyMat handleCvException (pure $ unsafeCoerceMat dst) $ withArrayPtr srcVec $ \srcVecPtr -> withPtr dst $ \dstPtr -> [cvExcept| std::vector buffer( $(Mat * srcVecPtr) , $(Mat * srcVecPtr) + $(int32_t c'temporalWindowSize) ); cv::fastNlMeansDenoisingColoredMulti( buffer , *$(Mat * dstPtr) , $(int32_t c'imgToDenoiseIndex) , $(int32_t c'temporalWindowSize) , $(double c'h) , $(double c'hColor) , $(int32_t templateWindowSize) , $(int32_t searchWindowSize) ); |] where c'h = realToFrac h c'hColor = realToFrac hColor c'srcVecLength = fromIntegral $ V.length srcVec -- if it is not odd we drop the last image c'temporalWindowSize | c'srcVecLength `mod` 2 == 1 = c'srcVecLength | otherwise = c'srcVecLength - 1 c'imgToDenoiseIndex = (c'temporalWindowSize - 1) `div` 2 {- | Perform denoise_TVL1 Example: @ denoise_TVL1Img :: forall h w w2 c d . ( Mat (ShapeT [h, w]) ('S c) ('S d) ~ Lenna_512x512 , w2 ~ ((*) w 2) ) => Mat ('S ['S h, 'S w2]) ('S c) ('S d) denoise_TVL1Img = exceptError $ do denoised <- matChannelMapM (denoise_TVL1 2 50 . V.singleton) lenna_512x512 withMatM (Proxy :: Proxy [h, w2]) (Proxy :: Proxy c) (Proxy :: Proxy d) black $ \\imgM -> do matCopyToM imgM (V2 0 0) lenna_512x512 Nothing matCopyToM imgM (V2 w 0) denoised Nothing where w = fromInteger $ natVal (Proxy :: Proxy w) @ <> -} denoise_TVL1 :: Double -- ^ details more is more 2 -> Int32 -- ^ Number of iterations that the algorithm will run -> V.Vector (Mat ('S [h, w]) ('S 1) ('S Word8)) -- ^ Vector of odd number of input 8-bit 3-channel images. -> CvExcept (Mat ('S [h, w]) ('S 1) ('S Word8)) -- ^ Output image same size and type as input. denoise_TVL1 lambda niters srcVec = unsafeWrapException $ do dst <- newEmptyMat handleCvException (pure $ unsafeCoerceMat dst) $ withArrayPtr srcVec $ \srcVecPtr -> withPtr dst $ \dstPtr -> [cvExcept| std::vector buffer( $(Mat * srcVecPtr) , $(Mat * srcVecPtr) + $(int32_t c'srcVecLength) ); cv::denoise_TVL1( buffer , *$(Mat * dstPtr) , $(double c'lambda) , $(int32_t niters) ); |] where c'lambda = realToFrac lambda c'srcVecLength = fromIntegral $ V.length srcVec {- | Perform decolor Decolor a color image to a grayscale (1 channel) and a color boosted image (3 channel) Example: @ decolorImg :: forall h h2 w w2 c d . ( Mat (ShapeT [h, w]) ('S c) ('S d) ~ Bikes_512x341 , h2 ~ ((*) h 2) , w2 ~ ((*) w 2) ) => Mat ('S ['S h2, 'S w2]) ('S c) ('S d) decolorImg = exceptError $ do (bikesGray, boost) <- decolor bikes_512x341 colorGray <- cvtColor gray bgr bikesGray withMatM (Proxy :: Proxy [h2, w2]) (Proxy :: Proxy c) (Proxy :: Proxy d) white $ \\imgM -> do matCopyToM imgM (V2 0 0) bikes_512x341 Nothing matCopyToM imgM (V2 0 h) colorGray Nothing matCopyToM imgM (V2 w h) boost Nothing where w = fromInteger $ natVal (Proxy :: Proxy w) h = fromInteger $ natVal (Proxy :: Proxy h) @ <> -} decolor :: Mat ('S [h, w]) ('S 3) ('S Word8) -- ^ Input image. -> CvExcept (Mat ('S [h, w]) ('S 1) ('S Word8), Mat ('S [h, w]) ('S 3) ('S Word8)) -- ^ Output images. decolor src = unsafeWrapException $ do gray <- newEmptyMat boost <- newEmptyMat handleCvException (pure (unsafeCoerceMat gray, unsafeCoerceMat boost)) $ withPtr src $ \srcPtr -> withPtr gray $ \grayPtr -> withPtr boost $ \boostPtr -> [cvExcept| cv::decolor( *$(Mat * srcPtr) , *$(Mat * grayPtr) , *$(Mat * boostPtr) ); |]